McCormick taps IBM's AI to develop new flavors, food products

The collaboration gives McCormick a way to speed up product development and give its 500-plus, global team of researchers a way to stay in sync.

Global food brand harnesses IBM's AI to make the perfect dressing The collaboration gives McCormick a way to speed up product development and give its 500-plus, global team of researchers a way to stay in sync. Read more: https://zd.net/2SytfNz

For the average home cook, creating a new recipe can be a fun, simple experiment -- they may pull from their list of go-to ingredients and spices to create different combinations of flavors. Sometimes it's works, and sometimes it doesn't.

For product developers in the food and flavor industry, it's a lot more complicated. Let's say they're tasked with developing a seasoning for a specific dish, such as a rice bowl. They'll have to meet a variety of requirements related to factors like the price range, ingredients used, sustainability or healthiness. Working within those constraints, the developer picks a "base formula" to start with -- they'll usually have around 5,000 or 10,000 base formulas to choose from. After they've added to the base formula, they'll send their new recipes out for testing. 

executive guide

What is AI? Everything you need to know about Artificial Intelligence

A guide to artificial intelligence, from machine learning and general AI to neural networks.

Read More

For a skilled product developer, it can take on average about 50 tries to come up with a recipe that's ready for commercialization. For the less-skilled, it can take 100 to 150 iterations. And in the food and flavor business, it can take a product developer anywhere from 10 to 20 years to become very proficient at their job.

"The whole problem is deceptively hard," Dr. Richard Goodwin of IBM told ZDNet.

That's why McCormick & Company is working with IBM to speed things up. For the past four years, the global brand has had its best developers working with IBM researchers to build an AI-powered platform for exploring different flavor combinations and predicting the ones that will be a hit.

With the new platform, developers should only need somewhere between 30 to 50 tries, rather than 100 to 150, to find the right recipe. This is a significant improvement for McCormick, which has more than 500 scientists, technologists and support staff working in its 20 R&D labs across five continents. The platform -- called "One" because it can unify that geographically-dispersed team -- should launch in mid-2019.

"When you look at our organization, we have different levels of skill sets among our scientists and developers," Dr. Hamed Faridi, McCormick's chief science officer, said to ZDNet. The vision for the platform, he said, "The vision was that every member has access to the knowledge and insight of the entire 500-person team... so everyone can be as good as the best developer we have."

The One platform leverages hundreds of millions of data points across the areas of sensory science, consumer preference and flavor palettes. McCormick has more than 70 years in R&D experience and has been collecting these data points for more than 40 years. The vast data set can even help McCormick account for regional or seasonal differences.

From this data, the system is learning a few different things. First, it's learning about ingredient pairing and substitutions. It's also learning to predict successful flavors based on the "creativity level" that product developers are looking for. For instance, a more "creative" barbecue sauce may include Chinese Five Spice or some other unexpected flavor that the system has determined pairs well with the base product. Lastly, it's learning to predict how similar two formulas will taste.

So far, the platform has been used to develop one-dish recipe mix flavors, which can be used to season proteins and vegetables. The initial offerings include Tuscan Chicken, Bourbon Pork Tenderloin and New Orleans Sausage.

As principal researcher and manager of IBM's Computational Creativity Research Group, Goodwin said that creativity is a subfield within AI that's gaining more traction. He cited the AI-generated painting that sold at auction last year for nearly a half million dollars.

"For a long time people have been asking whether computers can be intelligent," Goodwin said. "We've been looking at whether computers can help people be more creative."

Whether in food, painting or another creative field, there are certain tenets to meet, Goodwin said. First, the finished product must be "fit for purpose." For example, a potato chip drenched in olive oil may or may not taste good -- but it'd be too soggy to fit its purpose as a potato chip. Next, the finished product must be novel.

Lastly, it must fit in with society's expectations. Goodwin explained, "Bringing into American cuisine elements of other cultures and cuisines makes sense now, whereas in the 1950s it didn't."

For McCormick, the ultimate goal is to create a flavor -- either for its consumer products or its B2B products -- that becomes iconic, like Oreo cookies or Philadelphia cream cheese. Faridi said the One platform could help them get there.

"What we are seeing here is the tip of the iceberg," he said. "Three to four years down the road, the system will be a lot more capable than it is today. There is a lot more to come and a lot more to learn."